VPS benchmark

Benchmark evidence per provider

The measured samples behind each review, with the source they came from. The numbers are taken from published benchmarks, vendor documentation and community reports rather than tests we run ourselves, so treat them as samples from one node and verify the plan you actually buy.

Affiliate disclosure: We may earn a commission from some provider links at no additional cost to you.

G
Benchmark evidence2026-08-03

GreenCloud benchmark samples

The most useful community evidence is not a single headline score. A LowEndTalk review reported E5-2698 v4 nodes running Docker workloads such as Mastodon, Nextcloud and Rails apps with CPU steal reported at zero during the author's use. A later community post reported Geekbench 6 scores of 1,452 single-core and 7,358 multi-core on a different GreenCloud VPS. Treat those as independent samples, not a guarantee: storage, CPU generation and oversubscription can change by node.

Read the full review
C
Benchmark evidence2026-08-03

CloudCone benchmark samples

Older LowEndTalk tests are useful as a warning about variance rather than as a current score: one 2-vCPU sample identified an Intel Xeon E312xx Sandy Bridge CPU, while another review described E5-2620 nodes and roughly 100 Mbps practical network limits on a particular VPS. A separate 2024 community report documented packet loss above 20% on one assigned IP. These are not representative measurements for every CloudCone node, but they explain why a quick ping, MTR and download test should be part of your acceptance checklist.

Read the full review
V
Benchmark evidence2026-08-03

Vultr benchmark samples

Independent coverage is consistent about the shape of the result rather than one universal number. VPSBenchmarks lists Vultr in many countries and reports recent 2026 trials, while a 2026 independent review notes that High Frequency plans usually deliver tighter CPU and disk consistency than standard shared compute. Older Centmin Mod community tests also found location-level differences in storage I/O. The practical conclusion is simple: compare instance families and regions, not just the Vultr brand name.

Read the full review
H
Benchmark evidence2026-08-03

Hostinger benchmark samples

Published third-party measurements should be read with scope in mind. TechRadar's 2026 review measured a shared-hosting test site at 99.96% uptime, 382 ms average response time, 878 ms page load in its multi-location test and about 15 requests per second under a 20-user k6 test; those figures are not VPS benchmarks. VPSBenchmarks also lists Hostinger KVM trials using YABS, and community posts show why CPU steal should be monitored on the exact node. Use sysbench, fio, iperf3 and a small uptime monitor after deployment instead of copying a shared-plan score to a VPS claim.

Read the full review
R
Benchmark evidence2026-08-03

RackNerd benchmark samples

LowEndTalk benchmark threads show why RackNerd needs a plan-by-plan verdict. One 2024 Dublin sample reported Geekbench 6 scores of 343 single-core and 844 multi-core, while the same thread recorded roughly 1 GB/s sequential disk totals in its test. A 2026 Reddit comparison found different RackNerd boxes on similar plans using different Xeon generations, with one Dallas sample listed as an E5-2690 v4 scoring 933 single-core and 2,089 multi-core in YABS. These are useful field samples, not guaranteed specifications.

Read the full review

Want a plan matched to your workload instead of a benchmark score?

Size it with the recommender